A fracturing optimization identification method and system based on microseismic signal analysis

Abstract

The present application relates to the technical field of seismic signal analysis, in particular to a fracturing optimization identification method and system based on microseismic signal analysis; the method comprises: obtaining historical microseismic event data, including historical microseismic waveform data and historical fracturing pump pressure curve data and integrating to obtain an original data set; performing data preprocessing on the original data set to obtain denoised microseismic waveforms and smoothed pressure curves and performing simultaneous analysis to identify bridge plug setting signal characteristics and determine corresponding microseismic event types; based on the bridge plug setting signal characteristics in the microseismic event, an improved random forest model is used to identify and obtain microseismic events with classification labels. The present application introduces hierarchical heterogeneous tree groups by improving the random forest model, enhances the comprehensiveness and reliability of classification, removes redundant features, generates second-order feature interaction terms and weightedly calculates output probability, and improves the identification accuracy of the model for complex microseismic events.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of seismic signal analysis, and particularly relates to a fracturing optimization identification method and system based on microseismic signal analysis. BACKGROUND

[0002] Microseismic signal analysis is a technology for monitoring microseismic signals caused by fracturing in oil and gas exploitation and inverting fracture development characteristics; fracturing operation is a key engineering for producing fractures in reservoirs through pumping high-pressure fluid to improve oil and gas recovery.

[0003] In the existing fracturing optimization identification method based on microseismic signal analysis, the following problems exist: bridge plug setting signal identification relies on a single waveform or pressure curve, simultaneous analysis is not performed, and feature matching relies on an empirical threshold, resulting in missed or misjudged key events; redundant features are not effectively removed in feature processing, and second-order feature interaction analysis is lacking, affecting the quality of model input; traditional random forests are used in the classification model, and a single tree structure cannot balance local details and global correlation, resulting in insufficient classification accuracy and reliability, and the method cannot meet the needs of microseismic event accurate identification for fracturing optimization, and therefore, there is an urgent need for a fracturing optimization identification method based on microseismic signal analysis that can solve the above problems. SUMMARY

[0004] The present application aims to solve the problems in the background art and provides a fracturing optimization identification method and system based on microseismic signal analysis.

[0005] The technical solution of the present application is a fracturing optimization identification method based on microseismic signal analysis, comprising the following steps:

[0006] S1, obtaining historical microseismic event data, including historical microseismic waveform data and historical pressure curve data of a fracturing pump, and integrating the data to obtain an original data set;

[0007] S2, performing data preprocessing on the original data set to obtain denoised microseismic waveforms and smoothed pressure curves, and performing simultaneous analysis, positioning a fracturing interval period based on the pressure curve as a bridge plug setting candidate window, extracting waveform feature parameters in the window, and matching the features through a historical template library to identify bridge plug setting signal features and determine corresponding microseismic event types;

[0008] S3, based on the bridge plug setting signal features in the microseismic event, using an improved random forest model to identify microseismic events with classification labels;

[0009] S3 comprises:

[0010] S31, performing parallel classification using a layered heterogeneous tree group, including constructing a fine identification tree and a global correlation tree respectively according to a pre-allocated number of microseismic events as a training sample number, and obtaining classification probabilities output by each tree;

[0011] S32, calculating a classification label of a microseismic event by weighting the output probabilities according to a pre-allocated training sample number ratio.

[0012] Preferably, the simultaneous analysis method of S2 comprises:

[0013] S21, identifying an interval period from a fracturing end to a next fracturing start according to a smoothed pressure curve, and marking the interval period as a bridge plug setting candidate window;

[0014] S22, obtaining a denoised microseismic waveform of the bridge plug setting candidate window, calculating and obtaining waveform feature parameters to obtain candidate signal features; wherein the waveform feature parameters include rise time, peak amplitude, duration, main frequency, frequency band width, polarization direction angle and energy decay rate;

[0015] S23, obtaining a setting signal of a historical bridge plug of the same type, establishing a bridge plug setting signal library, and summarizing to obtain a template feature;

[0016] S24, calculating a feature matching degree of the candidate signal feature and the template feature, obtaining a qualified candidate signal feature based on the feature matching degree, and marking a bridge plug setting signal;

[0017] S25, determining a microseismic event according to the bridge plug setting signal and taking it as a simultaneous analysis result.

[0018] Preferably, S31 comprises:

[0019] S311, classifying features according to feature types to obtain a plurality of same-type feature lists based on the feature types;

[0020] S312, in the same-type feature list, obtaining all feature combinations by traversal, constructing a plurality of feature pairs, calculating Spearman correlation coefficients between the feature pairs, and marking feature pairs with Spearman correlation coefficients greater than a pre-set redundancy threshold as redundant feature pairs;

[0021] S313, obtaining an influence weight of a feature, sorting two features in the redundant feature pair according to the influence weight, marking the feature with the lowest sorting as a pruning feature, and excluding the pruning feature in the same-type feature list and updating the same-type feature list;

[0022] S314, determining core features of a microseismic event according to features in a plurality of updated same-type feature lists, obtaining all core feature combinations by traversal, and generating second-order core feature interaction terms;

[0023] S315, taking the interaction item and the core feature as the target item, calculating information gain of the target item and the event type, and screening high-differentiation target items.

[0024] Preferably, S31 further comprises:

[0025] S316, calculating information gain of the target item and the microseismic event type by the following formula:

[0026] ;

[0027] ;

[0028] ;

[0029] In the formula, H(C) is the basic information entropy; Ci is the microseismic event type; i is the event type number; i=(1, 2, 3); wherein, C1 is the artificial joint type; C2 is the natural joint type; C3 is the interference type; is the prior probability of the i-th microseismic event; N0 is the number of historical microseismic event samples; k is the number of possible value combination of the two core features in the target item X in the historical microseismic event samples; j is the possible value combination number; Nj is the sample number of the target item X in the j-th value of the k possible values; is the conditional entropy of the target item X under the event C; is the posterior probability of the i-th type; IG(X) is the information gain value of the target item X.

[0030] Preferably, S31 further comprises:

[0031] S317, screening the information gain value IG(X) of the target item X, retaining the target item with the information gain value IG(X)> the preset gain threshold value and taking it as the high-differentiation target item;

[0032] S318, calculating the dynamic weight W(X) of the retained target item X by the formula ;

[0033] In the formula, and are the minimum information gain value and the maximum information gain value respectively; G(X) is the adaptation coefficient of the target item X and the bottom lithology, which is determined based on industry experimental data; and a is a balance coefficient.

[0034] Preferably, S31 further comprises:

[0035] S319, obtain the target item value of the microseismic event and the corresponding microseismic event classification label and classification accuracy, and take the target item value with a classification accuracy greater than a preset accuracy threshold as a tree node;

[0036] S3110, sort the information gain values in descending order to obtain an information gain value order, and determine the target items of each layer according to the information gain value order and take the target items as a fine identification tree structure for depth splitting.

[0037] Preferably, S31 further comprises:

[0038] S3111, all target item combinations are traversed to construct a plurality of target item pairs, and for each target item pair, the global correlation score of the two target items in the target item pair is calculated through the following formula:

[0039]

[0040]

[0041] In the formula, MI is the mutual information value of the target item pair; Zh is the value of the target item pair, h is the value number of the target item pair, and h is a positive integer; Score is the global correlation score; and Acc is the classification accuracy of the target item pair in the historical sample;

[0042] S3112, the global correlation of the target item pair is sorted in descending order to obtain a global correlation descending order, and each layer of the target item pair is determined according to the global correlation descending order and taken as a global correlation tree structure for breadth splitting.

[0043] Preferably, S31 further comprises:

[0044] S3113, the node path of each tree is simulated, and after reaching the leaf node, the proportion of each type of microseismic event in the historical training sample of the node is counted and taken as the output classification probability.

[0045] The application further discloses a fracturing optimization identification system based on microseismic signal analysis, which applies the fracturing optimization identification method based on microseismic signal analysis.

[0046] The data acquisition module is used for obtaining historical microseismic event data, integrating historical microseismic waveform data and historical pressure curve data of a fracturing pump to obtain an original data set;

[0047] The data processing and analysis module is used for carrying out data preprocessing on the original data set to obtain denoised microseismic waveforms and smooth pressure curves and perform simultaneous analysis, identify bridge plug setting signal characteristics, and determine the corresponding microseismic event type.

[0048] ​​The signal recognition and event output module is used to identify and obtain microseismic events with classification labels based on the bridge plug setting signal characteristics in microseismic events using an improved random forest model.

[0049] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0050] The present invention uses a joint analysis of pressure curves and microseismic waveforms, combined with feature matching criteria, to accurately identify bridge plug setting signals, thereby improving the pertinence of key event feature extraction; by introducing a hierarchical heterogeneous tree group through an improved random forest model, parallel classification is achieved by constructing a fine recognition tree based on deep splitting of information gain and a global correlation tree based on global correlation breadth splitting of mutual information and accuracy, thereby enhancing the comprehensiveness and reliability of classification; by eliminating redundant features through the Spearman correlation coefficient, generating second-order feature interaction terms and weightedly calculating the output probability, the model's recognition accuracy for complex microseismic events is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of the first embodiment of the present invention. DETAILED DESCRIPTION

[0052] Example 1, as Figure 1 As shown, the present invention proposes a fracturing optimization identification method based on microseismic signal analysis, which includes the following steps:

[0053] S1. Obtain historical microseismic waveform data, historical fracturing pump pressure curve data, and classification labels and classification accuracy of microseismic types, and integrate them to obtain an original data set; wherein the microseismic waveform can be obtained by high-frequency sampling and recording of a microseismic acquisition instrument;

[0054] S2. Preprocess the original data set to obtain denoised microseismic waveforms and smoothed pressure curves, perform joint analysis, identify the characteristics of the bridge plug setting signal, and determine the corresponding microseismic event type;

[0055] The steps for data preprocessing of the original dataset include:

[0056] Taking the start time of the fracturing pump as the standard time, all data timestamps are uniformly converted into the time difference relative to the standard time;

[0057] The data with different sampling rates in the original dataset are resampled to the same time grid using cubic spline interpolation;

[0058] The correlation between the pressure curve and the microseismic signal is used to eliminate interference signals without pressure response. The wavelet threshold denoising algorithm is used to denoise the microseismic waveform to obtain a processed data set, including the denoised microseismic waveform and the smoothed pressure curve.

[0059] The method of simultaneous analysis of S2 comprises:

[0060] S21, identifying an interval period from the end of fracturing to the start of the next stage of fracturing according to the smooth pressure curve, and marking the interval period as a bridge plug setting candidate window;

[0061] S22, obtaining a denoised microseismic waveform of the bridge plug setting candidate window, calculating the waveform feature parameters to obtain the candidate signal features; wherein the waveform feature parameters include rise time, peak amplitude, duration, dominant frequency, frequency bandwidth, polarization direction angle and energy decay rate;

[0062] S23, obtaining the setting signals of the historical same type bridge plugs, establishing a bridge plug setting signal library, and summarizing the template features;

[0063] S24, calculating the feature matching degree of the candidate signal features and the template features, obtaining the qualified candidate signal features based on the feature matching degree and marking the bridge plug setting signal; wherein the feature matching degree is obtained by calculating the cosine similarity, and when the cosine similarity is greater than or equal to 0.85, it is determined that the feature matching degree is qualified;

[0064] The bridge plug setting signal excludes interference signals, including fracturing equipment operation interference signals and natural fracture random activation signals; for example, the fracturing equipment operation interference signal such as pump truck start-stop signal has a duration of 5-10 seconds and no sudden rise in pressure characteristics; the natural fracture random activation signal has no fixed time window, and the pressure curve has no synchronous sudden rise, and the energy is dispersed;

[0065] S25, determining a microseismic event according to the bridge plug setting signal and taking it as the simultaneous analysis result;

[0066] S3, based on the bridge plug setting signal features in the microseismic event, an improved random forest model is used to identify and obtain microseismic events with classification labels; specifically including the following steps: S31, using hierarchical heterogeneous tree groups for parallel classification, including constructing a fine identification tree and a global correlation tree according to the pre-allocated number of microseismic events as the number of training samples, and obtaining the classification probability output by each tree; specifically including the following steps:

[0067] S311, classifying the features according to the feature types to obtain a plurality of same type feature lists based on the feature types;

[0068] S312, in the same type feature list, all feature combinations are obtained by traversal, a plurality of feature pairs are constructed, the Spearman correlation coefficient between the feature pairs is calculated, and the feature pairs with a Spearman correlation coefficient greater than a preset redundancy threshold are marked as redundant feature pairs; wherein the preset redundancy threshold is determined based on historical microseismic big data;

[0069] S313, obtain the influence weight of the feature, sort the two features in the redundant feature pair according to the influence weight, mark the feature with the lowest sorting as a pruning feature, and remove the pruning feature from the same type feature list and update the same type feature list; wherein the influence weight of the feature can be obtained by fitting the recognition accuracy of the feature and the microseismic event in the historical microseismic event; or it can be obtained according to the importance score statistics of experts;

[0070] S314, determining the core feature of the microseismic event according to the features in the several updated same type feature lists, traversing to obtain all core feature combinations, and generating second-order core feature interaction terms;

[0071] S315, taking the interaction term and the core feature as target items, calculating the information gain of the target items and the event type, and screening high-discrimination target items;

[0072] S316, the information gain of the target item and the microseismic event type is calculated by the following formula:

[0073] ;

[0074] ;

[0075] ;

[0076] In the formula, H(C) is the basic information entropy; Ci is the microseismic event type; i is the event type number; i=(1, 2, 3); wherein C1 is the artificial joint type; C2 is the natural joint type; C3 is the interference type; is the prior probability of the i-th microseismic event; N0 is the number of historical microseismic event samples; k is the number of possible value combinations of two core features in the target item X in the historical microseismic event samples; j is the possible value combination number; Nj is the sample number of the target item X in the j-th value in k possible values; is the conditional entropy of the target item X under the event C; is the posterior probability of the i-th type; IG(X) is the information gain value of the target item X; wherein the larger the information gain value IG(X) is, the stronger the effect of the target item X on reducing the uncertainty of event classification is;

[0077] S317, screening the information gain value IG(X) of the target item X, retaining the target item with the information gain value IG(X)> preset gain threshold value and taking it as a high-discrimination target item; wherein the preset gain threshold value is obtained based on historical screening big data training;

[0078] S318, by formula The dynamic weight W(X) of the high-differentiation target item X is calculated;

[0079] In the formula, and respectively are the minimum value of information gain and the maximum value of information gain; G(X) is an adaptation coefficient of the target item X and the bottom lithology, which is determined based on industry experimental data; and a is a balance coefficient, which is used to preferentially ensure the dominant role of the information gain value;

[0080] S319, target item values of microseismic events and corresponding microseismic event classification labels and classification accuracies are obtained, and a target item value with a classification accuracy greater than a preset accuracy threshold is taken as a tree node; the preset accuracy threshold can be 0.85;

[0081] S3110, information gain values of target items are sorted in descending order to obtain information gain value sorting, and target items of each layer are determined according to the information gain value sorting and taken as a fine identification tree structure for depth splitting;

[0082] For example, first layer splitting: taking whether the main frequency is greater than 300 Hz as a node, high main frequency events such as artificial joints are preliminarily separated from low main frequency events such as natural joints or interference;

[0083] Second layer splitting: the second layer is split according to whether the polarization angle is <30°, to further focus on the directional features of artificial joints;

[0084] Third layer splitting: the third layer is split according to whether the amplitude attenuation rate is >0.7, to distinguish between natural joint amplitude attenuation and interference signal amplitude attenuation irregularity;

[0085] S3111, all target item combinations are traversed to construct a plurality of target item pairs, for each target item pair, the global correlation score of the two target items in the target item pair is calculated through the following formula:

[0086]

[0087]

[0088] In the formula, MI is the mutual information value of the target item pair; Zh is the target item pair value, h is the value number of the target item pair, and h is a positive integer; Score is the global correlation score; and Acc is the classification accuracy of the target item pair in historical samples;

[0089] S3112, the target item pairs are sorted in descending order of global correlation to obtain a global correlation descending order, and each layer of target item pairs is determined according to the global correlation descending order and taken as a global correlation tree structure for breadth splitting;

[0090] ​​S3113, node path simulation is performed on each tree, after reaching the leaf node, the proportion of each type of microseismic event in the historical training sample of the node is counted and used as the output classification probability.

[0091] S32, the classification label of the microseismic event is obtained by weighting the output probability according to the pre-allocated training sample quantity ratio; for example, the training sample quantity ratio used by the fine identification tree and the global correlation tree is 6 to 4, and the classification probabilities output by the two trees are weighted by 0.6 and 0.4 respectively.

[0092] By analyzing the pressure curve and the microseismic waveform together, and combining the feature matching criterion of cosine similarity ≥ 0.85, the bridge plug setting signal is accurately identified, and the pertinence of key event feature extraction is improved; by improving the random forest model and introducing a hierarchical heterogeneous tree group, parallel classification is realized by constructing a fine identification tree based on information gain depth splitting and a global correlation tree based on mutual information and accuracy, and the comprehensiveness and reliability of classification are enhanced; by eliminating redundant features through Spearman correlation coefficient, generating second-order feature interaction items and weighted computing output probability, the recognition accuracy of the model for complex microseismic events is improved.

[0093] In embodiment two, the present application proposes a fracturing optimization identification system based on microseismic signal analysis, which is applied to the fracturing optimization identification method based on microseismic signal analysis proposed in embodiment one, and specifically includes:

[0094] A data acquisition module is used to obtain historical microseismic event data, including historical microseismic waveform data and historical fracturing pump pressure curve data, and to integrate the data to obtain an original data set;

[0095] A data processing and analysis module is used to perform data preprocessing on the original data set to obtain denoised microseismic waveforms and smoothed pressure curves and perform joint analysis to identify bridge plug setting signal features and determine the corresponding microseismic event type;

[0096] A signal recognition and event output module is used to identify microseismic events with classification labels based on the bridge plug setting signal features in the microseismic events using an improved random forest model.

[0097] The embodiments of the present application are described in detail above in combination with the drawings, but the present application is not limited thereto, and various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the purpose of the present application.

Claims

1. A method for identifying fracturing optimization based on microseismic signal analysis, characterized in that, The method comprises the following steps: S1, obtaining historical microseismic event data, including historical microseismic waveform data and historical fracturing pump pressure curve data, and integrating to obtain an original data set; S2, performing data preprocessing on the original data set to obtain denoised microseismic waveforms and smoothed pressure curves, and performing simultaneous analysis, positioning a fracturing interval period based on the pressure curve as a bridge plug setting candidate window, extracting waveform feature parameters in the window, and matching the features through a historical template library to identify bridge plug setting signal characteristics and determine the corresponding microseismic event type; The simultaneous analysis method of S2 comprises: S21, identifying an interval period from the end of fracturing to the beginning of the next fracturing based on the smoothed pressure curve, and marking the interval period as a bridge plug setting candidate window; S22, obtaining denoised microseismic waveforms of the bridge plug setting candidate window, calculating waveform feature parameters to obtain candidate signal features; wherein the waveform feature parameters include rise time, peak amplitude, duration, dominant frequency, frequency band width, polarization direction angle and energy decay rate; S23, obtaining setting signals of historical bridge plugs of the same type, establishing a bridge plug setting signal library, and summarizing to obtain template features; S24, calculating the feature matching degree of the candidate signal features and the template features, obtaining qualified candidate signal features based on the feature matching degree, and marking the bridge plug setting signal; S25, determining a microseismic event based on the bridge plug setting signal and taking it as the simultaneous analysis result; S3, based on the bridge plug setting signal characteristics in the microseismic event, using an improved random forest model to identify and obtain microseismic events with classification labels; S3 comprises: S31, using a hierarchical heterogeneous tree group for parallel classification, including using the pre-allocated number of microseismic events as the number of training samples, respectively constructing a fine identification tree and a global correlation tree, and obtaining the classification probability output by each tree; S31 comprises: S311, classifying features according to feature types to obtain a plurality of same-type feature lists based on feature types; S312, in the same-type feature list, all feature combinations are obtained by traversal, a plurality of feature pairs are constructed, the Spearman correlation coefficient between the feature pairs is calculated, and the feature pairs with a Spearman correlation coefficient greater than a pre-set redundancy threshold are marked as redundant feature pairs; S313, obtaining the influence weight of the features, sorting the two features in the redundant feature pair according to the influence weight, marking the feature with the lowest sorting as a pruning feature, and deleting the pruning feature in the same-type feature list and updating the same-type feature list; S314, determining the core features of the microseismic event according to the features in the plurality of updated same-type feature lists, obtaining all core feature combinations by traversal, and generating second-order core feature interaction terms; S315, taking the interaction terms and the core features as target items, calculating the information gain of the target items and the event type, and screening high-discrimination target items; S32, weighting and calculating the output probability according to the pre-allocated training sample quantity ratio to obtain the classification label of the microseismic event.

2. The method of claim 1, wherein, S31 further comprises: S316, calculating the information gain of the target items and the microseismic event type by the following formula: ； ； ； In the formula, H(C) is the basic information entropy; Ci is the microseismic event type; i is the event type number; i=(1, 2, 3); wherein, C1 is the artificial joint type; C2 is the natural joint type; C3 is the interference type; is the prior probability of the i-th microseismic event; N0 is the number of historical microseismic event samples; k is the number of possible value combination of two core features of the target item X in the historical microseismic event samples; j is the possible value combination number; Nj is the sample number of the target item X when the j-th value in the k possible values; is the conditional entropy of the target item X under the event C; is the posterior probability of the i-th type; IG(X) is the information gain value of the target item X.

3. The method of claim 2, wherein, S31 further comprises: S317, screening the information gain value IG(X) of the target item X, retaining the target item with the information gain value IG(X) greater than a preset gain threshold value as a high-differentiation target item; S318、 through the formula The dynamic weight W(X) of the target item X reserved is calculated. In the formula, and respectively are the minimum and maximum values of information gain; G(X) is the fitting coefficient of target item X and the underlying lithology, which is determined based on industry experimental data; and a is a balance coefficient.

4. The method of claim 3, wherein, S31 further includes: S319, obtaining the target item value of the microseismic event and the corresponding microseismic event classification label and classification accuracy, and taking the target item value with the classification accuracy greater than a preset accuracy threshold value as a tree node; S3110, performing descending order sorting according to the information gain value of the target item to obtain an information gain value sorting, and determining the target item of each layer according to the information gain value sorting and taking the target item as a fine recognition tree structure for depth splitting.

5. The method of claim 4, wherein, S31 further includes: S3111, traversing all target item combinations to construct a plurality of target item pairs, and for each target item pair, calculating the global correlation score of the two target items in the target item pair by the following formula: ； ； In the formula, MI is the mutual information value of the target item pair; Zh is the value of the target item pair, h is the value number of the target item pair, and h is a positive integer; Score is the global correlation score; and Acc is the classification accuracy of the target item pair in the historical sample; S3112, performing descending order sorting according to the global correlation of the target item pair to obtain a global correlation descending order sorting, and determining the target item pair of each layer according to the global correlation descending order sorting and taking the target item pair as a global correlation tree structure for breadth splitting.

6. The method of claim 5, wherein, S31 further includes: S3113, performing node path simulation on each tree, and after reaching a leaf node, counting the proportion of each type of microseismic event in the historical training sample of the node and taking the proportion as the output classification probability.

7. A fracturing optimization identification system based on microseismic signal analysis, applied to the fracturing optimization identification method based on microseismic signal analysis in any one of claims 1 to 6, characterized in that, Specifically, it includes: A data acquisition module configured to obtain historical microseismic event data, including historical microseismic waveform data and historical fracturing pump pressure curve data, and integrate the data to obtain an original data set; A data processing and analysis module configured to perform data preprocessing on the original data set to obtain denoised microseismic waveforms and smoothed pressure curves, and perform joint analysis to identify a bridge plug setting signal feature and determine a corresponding microseismic event type; A signal recognition and event output module configured to identify a microseismic event with a classification label based on the bridge plug setting signal feature in the microseismic event by using an improved random forest model.

Images